Determining an Amount of Data of a Track to Stage into Cache Using a Machine Learning Module

1. A computer program product for determining data to stage into cache from a storage, wherein the computer program product comprises a computer readable storage medium having computer readable program code embodied therein that when executed performs operations, the operations comprising: providing performance attributes related to staging tracks from the storage to the cache to a machine learning module; receiving, from the machine learning module having processed the performance attributes, an output indicating a staging strategy indicating an amount of data of a track to stage into the cache comprising one of a plurality of staging strategies; and staging the amount of data, specified by the staging strategy indicated in the output, from an accessed track that is not in the cache.

2. The computer program product of claim 1 , wherein the amount of data of a track to stage as part of the plurality of staging strategies includes at least a plurality of a partial track staging to stage all sectors from a requested sector of a track, a sector staging to stage only the requested sectors of the track, and a full track staging to stage all sectors of the track.

3. The computer program product of claim 1 , wherein the performance attributes provided to the machine learning module comprise a plurality of: cache misses indicating a number of requests to access a track that is not in the cache; cache hits indicating a number of requests to access a track that is in the cache; front accesses incremented in response to a request to a track indicated as in the cache and the request having a first requested sector of at least one sector requested that precedes a first staged sector of the track; back accesses incremented in response to a request to a track indicated as in the cache and the request having a first requested sector of at least one sector requested that follows a last staged sector of the track; speed of the storage including a specified speed of at least one storage device in which the storage is implemented; a response time to respond to requests for sectors; a current adaptor bandwidth through which data is transferred between the cache and the storage; and optimum adaptor bandwidth indicating for transferring data between the cache and the storage.

4. The computer program product of claim 1 , wherein the performance attributes include: historical front end access ratio comprising ((front accesses divided by (cache hits plus cache misses)) plus a previously calculated historical front end access ratio) divided by two; and historical alternate record access ratio comprising (((front accesses plus back accesses) divided by (cache hits plus cache misses)) plus a previously calculated historical alternate record access ratio) divided by two.

5. The computer program product of claim 1 , wherein the operations further comprise: receiving a request to access a requested track; and determining whether the requested track is in the cache, wherein the providing the performance attributes to the machine learning module is performed in response to determining the requested track is not in the cache.

6. The computer program product of claim 1 , wherein the operations further comprise: determining a predetermined number of cache misses, wherein the providing the performance attributes to the machine learning module is performed after the predetermined number of cache misses.

7. The computer program product of claim 1 , wherein the performance attributes include at least a plurality of cache misses indicating a number of requests to access a track that is not in the cache, cache hits indicating a number of requests to access a track that is in the cache, front accesses incremented in response to a request to a track indicated as in the cache and the request having a first requested sector of at least one sector requested that precedes a first staged sector of the track, and back accesses incremented in response to a request to a track indicated as in the cache and the request having a first requested sector of at least one sector requested that follows a last staged sector of the track, wherein the operations further comprise: incrementing cache hits in response to determining that the requested track is in the cache; incrementing the cache misses in response to determining that the requested track is not in the cache; determining whether a first of requested sectors in the requested track are in the cache in response to determining that the requested track is in the cache; incrementing the back accesses in response to determining that the first of requested sectors follow sectors of the track in the cache; and incrementing the front accesses in response to determining that the first of requested sectors precedes sectors of the track in the cache.

8. A system for determining data to stage from a storage, comprising: a processor; a cache; and a computer readable storage medium having computer readable program code embodied therein that when executed performs operations, the operations comprising: providing performance attributes related to staging tracks from the storage to the cache to a machine learning module; receiving, from the machine learning module having processed the performance attributes, an output indicating a staging strategy indicating an amount of data of a track to stage into the cache comprising one of a plurality of staging strategies; and staging the amount of data, specified by the staging strategy indicated in the output, from an accessed track that is not in the cache.

9. The system of claim 8 , wherein the amount of data of a track to stage as part of the plurality of staging strategies includes at least a plurality of a partial track staging to stage all sectors from a requested sector of a track, a sector staging to stage only the requested sectors of the track, and a full track staging to stage all sectors of the track.

10. The system of claim 8 , wherein the performance attributes provided to the machine learning module comprise a plurality of: cache misses indicating a number of requests to access a track that is not in the cache; cache hits indicating a number of requests to access a track that is in the cache; front accesses incremented in response to a request to a track indicated as in the cache and the request having a first requested sector of at least one sector requested that precedes a first staged sector of the track; back accesses incremented in response to a request to a track indicated as in the cache and the request having a first requested sector of at least one sector requested that follows a last staged sector of the track; speed of the storage including a specified speed of at least one storage device in which the storage is implemented; a response time to respond to requests for sectors; a current adaptor bandwidth through which data is transferred between the cache and the storage; and optimum adaptor bandwidth indicating for transferring data between the cache and the storage.

11. The system of claim 8 , wherein the performance attributes include: historical front end access ratio comprising ((front accesses divided by (cache hits plus cache misses)) plus a previously calculated historical front end access ratio) divided by two; and historical alternate record access ratio comprising (((front accesses plus back accesses) divided by (cache hits plus cache misses)) plus a previously calculated historical alternate record access ratio) divided by two.

12. The system of claim 8 , wherein the operations further comprise: receiving a request to access a requested track; and determining whether the requested track is in the cache, wherein the providing the performance attributes to the machine learning module is performed in response to determining the requested track is not in the cache.

13. The system of claim 8 , wherein the operations further comprise: determining a predetermined number of cache misses, wherein the providing the performance attributes to the machine learning module is performed after the predetermined number of cache misses.

14. The system of claim 8 , wherein the performance attributes include at least a plurality of cache misses indicating a number of requests to access a track that is not in the cache, cache hits indicating a number of requests to access a track that is in the cache, front accesses incremented in response to a request to a track indicated as in the cache and the request having a first requested sector of at least one sector requested that precedes a first staged sector of the track, and back accesses incremented in response to a request to a track indicated as in the cache and the request having a first requested sector of at least one sector requested that follows a last staged sector of the track, wherein the operations further comprise: incrementing cache hits in response to determining that the requested track is in the cache; incrementing the cache misses in response to determining that the requested track is not in the cache; determining whether a first of requested sectors in the requested track are in the cache in response to determining that the requested track is in the cache; incrementing the back accesses in response to determining that the first of requested sectors follow sectors of the track in the cache; and incrementing the front accesses in response to determining that the first of requested sectors precedes sectors of the track in the cache.

15. A method for determining data to stage into cache from a storage, comprising: providing performance attributes related to staging tracks from the storage to the cache to a machine learning module; receiving, from the machine learning module having processed the performance attributes, an output indicating a staging strategy indicating an amount of data of a track to stage into the cache comprising one of a plurality of staging strategies; and staging the amount of data, specified by the staging strategy indicated in the output, from an accessed track that is not in the cache.

16. The method of claim 15 , wherein the amount of data of a track to stage as part of the plurality of staging strategies includes at least a plurality of a partial track staging to stage all sectors from a requested sector of a track, a sector staging to stage only the requested sectors of the track, and a full track staging to stage all sectors of the track.

17. The method of claim 15 , wherein the performance attributes provided to the machine learning module comprise a plurality of: cache misses indicating a number of requests to access a track that is not in the cache; cache hits indicating a number of requests to access a track that is in the cache; front accesses incremented in response to a request to a track indicated as in the cache and the request having a first requested sector of at least one sector requested that precedes a first staged sector of the track; back accesses incremented in response to a request to a track indicated as in the cache and the request having a first requested sector of at least one sector requested that follows a last staged sector of the track; speed of the storage including a specified speed of at least one storage device in which the storage is implemented; a response time to respond to requests for sectors; a current adaptor bandwidth through which data is transferred between the cache and the storage; and optimum adaptor bandwidth indicating for transferring data between the cache and the storage.

18. The method of claim 15 , wherein the performance attributes include: historical front end access ratio comprising ((front accesses divided by (cache hits plus cache misses)) plus a previously calculated historical front end access ratio) divided by two; and historical alternate record access ratio comprising (((front accesses plus back accesses) divided by (cache hits plus cache misses)) plus a previously calculated historical alternate record access ratio) divided by two.

19. The method of claim 15 , further comprising: receiving a request to access a requested track; and determining whether the requested track is in the cache, wherein the providing the performance attributes to the machine learning module is performed in response to determining the requested track is not in the cache.

20. The method of claim 15 , further comprising: determining a predetermined number of cache misses, wherein the providing the performance attributes to the machine learning module is performed after the predetermined number of cache misses.